Illustrative Photo
Meta has released an “open source” version of Google's popular NotebookLM podcasting feature. A project called NotebookLlama uses Meta's own Llama models for most of the processing.
TechCrunch writes about it.
First NotebookLlama creates a transcript from file – for example, a PDF file of a news or blog post. It then adds “more drama” and pauses for more realism before feeding the transcript into open text-to-speech models. The text is then fed into open speech synthesis models to produce an audio version.
200% Deposit Bonus up to €3,000 180% First Deposit Bonus up to $20,000The results don't sound as good as NotebookLM, users say. The voices have a very obvious robotic quality and a tendency to shout over each other in obscure places. But the Meta researchers behind the project say the quality could be improved with stronger models.
The text-to-speech model is a limit to how natural it will sound, the Meta researchers write on the NotebookLlama page on GitHub. Another approach to writing a podcast is for two agents to discuss a topic of interest and write an outline for the podcast. We now use a single model for writing a podcast outline.
NotebookLlama